drivers/md/bcache/sysfs.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * bcache sysfs interfaces
   4  *
   5  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
   6  * Copyright 2012 Google, Inc.
   7  */
   8
   9 #include "bcache.h"
  10 #include "sysfs.h"
  11 #include "btree.h"
  12 #include "request.h"
  13 #include "writeback.h"
  14
  15 #include <linux/blkdev.h>
  16 #include <linux/sort.h>
  17 #include <linux/sched/clock.h>
  18
  19 static const char * const cache_replacement_policies[] = {
  20         "lru",
  21         "fifo",
  22         "random",
  23         NULL
  24 };
  25
  26 static const char * const error_actions[] = {
  27         "unregister",
  28         "panic",
  29         NULL
  30 };
  31
  32 write_attribute(attach);
  33 write_attribute(detach);
  34 write_attribute(unregister);
  35 write_attribute(stop);
  36 write_attribute(clear_stats);
  37 write_attribute(trigger_gc);
  38 write_attribute(prune_cache);
  39 write_attribute(flash_vol_create);
  40
  41 read_attribute(bucket_size);
  42 read_attribute(block_size);
  43 read_attribute(nbuckets);
  44 read_attribute(tree_depth);
  45 read_attribute(root_usage_percent);
  46 read_attribute(priority_stats);
  47 read_attribute(btree_cache_size);
  48 read_attribute(btree_cache_max_chain);
  49 read_attribute(cache_available_percent);
  50 read_attribute(written);
  51 read_attribute(btree_written);
  52 read_attribute(metadata_written);
  53 read_attribute(active_journal_entries);
  54
  55 sysfs_time_stats_attribute(btree_gc,    sec, ms);
  56 sysfs_time_stats_attribute(btree_split, sec, us);
  57 sysfs_time_stats_attribute(btree_sort,  ms,  us);
  58 sysfs_time_stats_attribute(btree_read,  ms,  us);
  59
  60 read_attribute(btree_nodes);
  61 read_attribute(btree_used_percent);
  62 read_attribute(average_key_size);
  63 read_attribute(dirty_data);
  64 read_attribute(bset_tree_stats);
  65
  66 read_attribute(state);
  67 read_attribute(cache_read_races);
  68 read_attribute(writeback_keys_done);
  69 read_attribute(writeback_keys_failed);
  70 read_attribute(io_errors);
  71 read_attribute(congested);
  72 rw_attribute(congested_read_threshold_us);
  73 rw_attribute(congested_write_threshold_us);
  74
  75 rw_attribute(sequential_cutoff);
  76 rw_attribute(data_csum);
  77 rw_attribute(cache_mode);
  78 rw_attribute(writeback_metadata);
  79 rw_attribute(writeback_running);
  80 rw_attribute(writeback_percent);
  81 rw_attribute(writeback_delay);
  82 rw_attribute(writeback_rate);
  83
  84 rw_attribute(writeback_rate_update_seconds);
  85 rw_attribute(writeback_rate_d_term);
  86 rw_attribute(writeback_rate_p_term_inverse);
  87 read_attribute(writeback_rate_debug);
  88
  89 read_attribute(stripe_size);
  90 read_attribute(partial_stripes_expensive);
  91
  92 rw_attribute(synchronous);
  93 rw_attribute(journal_delay_ms);
  94 rw_attribute(discard);
  95 rw_attribute(running);
  96 rw_attribute(label);
  97 rw_attribute(readahead);
  98 rw_attribute(errors);
  99 rw_attribute(io_error_limit);
 100 rw_attribute(io_error_halflife);
 101 rw_attribute(verify);
 102 rw_attribute(bypass_torture_test);
 103 rw_attribute(key_merging_disabled);
 104 rw_attribute(gc_always_rewrite);
 105 rw_attribute(expensive_debug_checks);
 106 rw_attribute(cache_replacement_policy);
 107 rw_attribute(btree_shrinker_disabled);
 108 rw_attribute(copy_gc_enabled);
 109 rw_attribute(size);
 110
 111 SHOW(__bch_cached_dev)
 112 {
 113         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 114                                              disk.kobj);
 115         const char *states[] = { "no cache", "clean", "dirty", "inconsistent" };
 116
 117 #define var(stat)               (dc->stat)
 118
 119         if (attr == &sysfs_cache_mode)
 120                 return bch_snprint_string_list(buf, PAGE_SIZE,
 121                                                bch_cache_modes + 1,
 122                                                BDEV_CACHE_MODE(&dc->sb));
 123
 124         sysfs_printf(data_csum,         "%i", dc->disk.data_csum);
 125         var_printf(verify,              "%i");
 126         var_printf(bypass_torture_test, "%i");
 127         var_printf(writeback_metadata,  "%i");
 128         var_printf(writeback_running,   "%i");
 129         var_print(writeback_delay);
 130         var_print(writeback_percent);
 131         sysfs_hprint(writeback_rate,    dc->writeback_rate.rate << 9);
 132
 133         var_print(writeback_rate_update_seconds);
 134         var_print(writeback_rate_d_term);
 135         var_print(writeback_rate_p_term_inverse);
 136
 137         if (attr == &sysfs_writeback_rate_debug) {
 138                 char rate[20];
 139                 char dirty[20];
 140                 char target[20];
 141                 char proportional[20];
 142                 char derivative[20];
 143                 char change[20];
 144                 s64 next_io;
 145
 146                 bch_hprint(rate,        dc->writeback_rate.rate << 9);
 147                 bch_hprint(dirty,       bcache_dev_sectors_dirty(&dc->disk) << 9);
 148                 bch_hprint(target,      dc->writeback_rate_target << 9);
 149                 bch_hprint(proportional,dc->writeback_rate_proportional << 9);
 150                 bch_hprint(derivative,  dc->writeback_rate_derivative << 9);
 151                 bch_hprint(change,      dc->writeback_rate_change << 9);
 152
 153                 next_io = div64_s64(dc->writeback_rate.next - local_clock(),
 154                                     NSEC_PER_MSEC);
 155
 156                 return sprintf(buf,
 157                                "rate:\t\t%s/sec\n"
 158                                "dirty:\t\t%s\n"
 159                                "target:\t\t%s\n"
 160                                "proportional:\t%s\n"
 161                                "derivative:\t%s\n"
 162                                "change:\t\t%s/sec\n"
 163                                "next io:\t%llims\n",
 164                                rate, dirty, target, proportional,
 165                                derivative, change, next_io);
 166         }
 167
 168         sysfs_hprint(dirty_data,
 169                      bcache_dev_sectors_dirty(&dc->disk) << 9);
 170
 171         sysfs_hprint(stripe_size,       dc->disk.stripe_size << 9);
 172         var_printf(partial_stripes_expensive,   "%u");
 173
 174         var_hprint(sequential_cutoff);
 175         var_hprint(readahead);
 176
 177         sysfs_print(running,            atomic_read(&dc->running));
 178         sysfs_print(state,              states[BDEV_STATE(&dc->sb)]);
 179
 180         if (attr == &sysfs_label) {
 181                 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
 182                 buf[SB_LABEL_SIZE + 1] = '\0';
 183                 strcat(buf, "\n");
 184                 return strlen(buf);
 185         }
 186
 187 #undef var
 188         return 0;
 189 }
 190 SHOW_LOCKED(bch_cached_dev)
 191
 192 STORE(__cached_dev)
 193 {
 194         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 195                                              disk.kobj);
 196         ssize_t v;
 197         struct cache_set *c;
 198         struct kobj_uevent_env *env;
 199
 200 #define d_strtoul(var)          sysfs_strtoul(var, dc->var)
 201 #define d_strtoul_nonzero(var)  sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
 202 #define d_strtoi_h(var)         sysfs_hatoi(var, dc->var)
 203
 204         sysfs_strtoul(data_csum,        dc->disk.data_csum);
 205         d_strtoul(verify);
 206         d_strtoul(bypass_torture_test);
 207         d_strtoul(writeback_metadata);
 208         d_strtoul(writeback_running);
 209         d_strtoul(writeback_delay);
 210
 211         sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent, 0, 40);
 212
 213         sysfs_strtoul_clamp(writeback_rate,
 214                             dc->writeback_rate.rate, 1, INT_MAX);
 215
 216         d_strtoul_nonzero(writeback_rate_update_seconds);
 217         d_strtoul(writeback_rate_d_term);
 218         d_strtoul_nonzero(writeback_rate_p_term_inverse);
 219
 220         sysfs_strtoul_clamp(sequential_cutoff,
 221                             dc->sequential_cutoff,
 222                             0, UINT_MAX);
 223         d_strtoi_h(readahead);
 224
 225         if (attr == &sysfs_clear_stats)
 226                 bch_cache_accounting_clear(&dc->accounting);
 227
 228         if (attr == &sysfs_running &&
 229             strtoul_or_return(buf))
 230                 bch_cached_dev_run(dc);
 231
 232         if (attr == &sysfs_cache_mode) {
 233                 v = bch_read_string_list(buf, bch_cache_modes + 1);
 234
 235                 if (v < 0)
 236                         return v;
 237
 238                 if ((unsigned) v != BDEV_CACHE_MODE(&dc->sb)) {
 239                         SET_BDEV_CACHE_MODE(&dc->sb, v);
 240                         bch_write_bdev_super(dc, NULL);
 241                 }
 242         }
 243
 244         if (attr == &sysfs_label) {
 245                 if (size > SB_LABEL_SIZE)
 246                         return -EINVAL;
 247                 memcpy(dc->sb.label, buf, size);
 248                 if (size < SB_LABEL_SIZE)
 249                         dc->sb.label[size] = '\0';
 250                 if (size && dc->sb.label[size - 1] == '\n')
 251                         dc->sb.label[size - 1] = '\0';
 252                 bch_write_bdev_super(dc, NULL);
 253                 if (dc->disk.c) {
 254                         memcpy(dc->disk.c->uuids[dc->disk.id].label,
 255                                buf, SB_LABEL_SIZE);
 256                         bch_uuid_write(dc->disk.c);
 257                 }
 258                 env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
 259                 if (!env)
 260                         return -ENOMEM;
 261                 add_uevent_var(env, "DRIVER=bcache");
 262                 add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
 263                 add_uevent_var(env, "CACHED_LABEL=%s", buf);
 264                 kobject_uevent_env(
 265                         &disk_to_dev(dc->disk.disk)->kobj, KOBJ_CHANGE, env->envp);
 266                 kfree(env);
 267         }
 268
 269         if (attr == &sysfs_attach) {
 270                 uint8_t         set_uuid[16];
 271
 272                 if (bch_parse_uuid(buf, set_uuid) < 16)
 273                         return -EINVAL;
 274
 275                 v = -ENOENT;
 276                 list_for_each_entry(c, &bch_cache_sets, list) {
 277                         v = bch_cached_dev_attach(dc, c, set_uuid);
 278                         if (!v)
 279                                 return size;
 280                 }
 281
 282                 pr_err("Can't attach %s: cache set not found", buf);
 283                 return v;
 284         }
 285
 286         if (attr == &sysfs_detach && dc->disk.c)
 287                 bch_cached_dev_detach(dc);
 288
 289         if (attr == &sysfs_stop)
 290                 bcache_device_stop(&dc->disk);
 291
 292         return size;
 293 }
 294
 295 STORE(bch_cached_dev)
 296 {
 297         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 298                                              disk.kobj);
 299
 300         mutex_lock(&bch_register_lock);
 301         size = __cached_dev_store(kobj, attr, buf, size);
 302
 303         if (attr == &sysfs_writeback_running)
 304                 bch_writeback_queue(dc);
 305
 306         if (attr == &sysfs_writeback_percent)
 307                 schedule_delayed_work(&dc->writeback_rate_update,
 308                                       dc->writeback_rate_update_seconds * HZ);
 309
 310         mutex_unlock(&bch_register_lock);
 311         return size;
 312 }
 313
 314 static struct attribute *bch_cached_dev_files[] = {
 315         &sysfs_attach,
 316         &sysfs_detach,
 317         &sysfs_stop,
 318 #if 0
 319         &sysfs_data_csum,
 320 #endif
 321         &sysfs_cache_mode,
 322         &sysfs_writeback_metadata,
 323         &sysfs_writeback_running,
 324         &sysfs_writeback_delay,
 325         &sysfs_writeback_percent,
 326         &sysfs_writeback_rate,
 327         &sysfs_writeback_rate_update_seconds,
 328         &sysfs_writeback_rate_d_term,
 329         &sysfs_writeback_rate_p_term_inverse,
 330         &sysfs_writeback_rate_debug,
 331         &sysfs_dirty_data,
 332         &sysfs_stripe_size,
 333         &sysfs_partial_stripes_expensive,
 334         &sysfs_sequential_cutoff,
 335         &sysfs_clear_stats,
 336         &sysfs_running,
 337         &sysfs_state,
 338         &sysfs_label,
 339         &sysfs_readahead,
 340 #ifdef CONFIG_BCACHE_DEBUG
 341         &sysfs_verify,
 342         &sysfs_bypass_torture_test,
 343 #endif
 344         NULL
 345 };
 346 KTYPE(bch_cached_dev);
 347
 348 SHOW(bch_flash_dev)
 349 {
 350         struct bcache_device *d = container_of(kobj, struct bcache_device,
 351                                                kobj);
 352         struct uuid_entry *u = &d->c->uuids[d->id];
 353
 354         sysfs_printf(data_csum, "%i", d->data_csum);
 355         sysfs_hprint(size,      u->sectors << 9);
 356
 357         if (attr == &sysfs_label) {
 358                 memcpy(buf, u->label, SB_LABEL_SIZE);
 359                 buf[SB_LABEL_SIZE + 1] = '\0';
 360                 strcat(buf, "\n");
 361                 return strlen(buf);
 362         }
 363
 364         return 0;
 365 }
 366
 367 STORE(__bch_flash_dev)
 368 {
 369         struct bcache_device *d = container_of(kobj, struct bcache_device,
 370                                                kobj);
 371         struct uuid_entry *u = &d->c->uuids[d->id];
 372
 373         sysfs_strtoul(data_csum,        d->data_csum);
 374
 375         if (attr == &sysfs_size) {
 376                 uint64_t v;
 377                 strtoi_h_or_return(buf, v);
 378
 379                 u->sectors = v >> 9;
 380                 bch_uuid_write(d->c);
 381                 set_capacity(d->disk, u->sectors);
 382         }
 383
 384         if (attr == &sysfs_label) {
 385                 memcpy(u->label, buf, SB_LABEL_SIZE);
 386                 bch_uuid_write(d->c);
 387         }
 388
 389         if (attr == &sysfs_unregister) {
 390                 set_bit(BCACHE_DEV_DETACHING, &d->flags);
 391                 bcache_device_stop(d);
 392         }
 393
 394         return size;
 395 }
 396 STORE_LOCKED(bch_flash_dev)
 397
 398 static struct attribute *bch_flash_dev_files[] = {
 399         &sysfs_unregister,
 400 #if 0
 401         &sysfs_data_csum,
 402 #endif
 403         &sysfs_label,
 404         &sysfs_size,
 405         NULL
 406 };
 407 KTYPE(bch_flash_dev);
 408
 409 struct bset_stats_op {
 410         struct btree_op op;
 411         size_t nodes;
 412         struct bset_stats stats;
 413 };
 414
 415 static int bch_btree_bset_stats(struct btree_op *b_op, struct btree *b)
 416 {
 417         struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
 418
 419         op->nodes++;
 420         bch_btree_keys_stats(&b->keys, &op->stats);
 421
 422         return MAP_CONTINUE;
 423 }
 424
 425 static int bch_bset_print_stats(struct cache_set *c, char *buf)
 426 {
 427         struct bset_stats_op op;
 428         int ret;
 429
 430         memset(&op, 0, sizeof(op));
 431         bch_btree_op_init(&op.op, -1);
 432
 433         ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, bch_btree_bset_stats);
 434         if (ret < 0)
 435                 return ret;
 436
 437         return snprintf(buf, PAGE_SIZE,
 438                         "btree nodes:           %zu\n"
 439                         "written sets:          %zu\n"
 440                         "unwritten sets:                %zu\n"
 441                         "written key bytes:     %zu\n"
 442                         "unwritten key bytes:   %zu\n"
 443                         "floats:                        %zu\n"
 444                         "failed:                        %zu\n",
 445                         op.nodes,
 446                         op.stats.sets_written, op.stats.sets_unwritten,
 447                         op.stats.bytes_written, op.stats.bytes_unwritten,
 448                         op.stats.floats, op.stats.failed);
 449 }
 450
 451 static unsigned bch_root_usage(struct cache_set *c)
 452 {
 453         unsigned bytes = 0;
 454         struct bkey *k;
 455         struct btree *b;
 456         struct btree_iter iter;
 457
 458         goto lock_root;
 459
 460         do {
 461                 rw_unlock(false, b);
 462 lock_root:
 463                 b = c->root;
 464                 rw_lock(false, b, b->level);
 465         } while (b != c->root);
 466
 467         for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
 468                 bytes += bkey_bytes(k);
 469
 470         rw_unlock(false, b);
 471
 472         return (bytes * 100) / btree_bytes(c);
 473 }
 474
 475 static size_t bch_cache_size(struct cache_set *c)
 476 {
 477         size_t ret = 0;
 478         struct btree *b;
 479
 480         mutex_lock(&c->bucket_lock);
 481         list_for_each_entry(b, &c->btree_cache, list)
 482                 ret += 1 << (b->keys.page_order + PAGE_SHIFT);
 483
 484         mutex_unlock(&c->bucket_lock);
 485         return ret;
 486 }
 487
 488 static unsigned bch_cache_max_chain(struct cache_set *c)
 489 {
 490         unsigned ret = 0;
 491         struct hlist_head *h;
 492
 493         mutex_lock(&c->bucket_lock);
 494
 495         for (h = c->bucket_hash;
 496              h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
 497              h++) {
 498                 unsigned i = 0;
 499                 struct hlist_node *p;
 500
 501                 hlist_for_each(p, h)
 502                         i++;
 503
 504                 ret = max(ret, i);
 505         }
 506
 507         mutex_unlock(&c->bucket_lock);
 508         return ret;
 509 }
 510
 511 static unsigned bch_btree_used(struct cache_set *c)
 512 {
 513         return div64_u64(c->gc_stats.key_bytes * 100,
 514                          (c->gc_stats.nodes ?: 1) * btree_bytes(c));
 515 }
 516
 517 static unsigned bch_average_key_size(struct cache_set *c)
 518 {
 519         return c->gc_stats.nkeys
 520                 ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
 521                 : 0;
 522 }
 523
 524 SHOW(__bch_cache_set)
 525 {
 526         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 527
 528         sysfs_print(synchronous,                CACHE_SYNC(&c->sb));
 529         sysfs_print(journal_delay_ms,           c->journal_delay_ms);
 530         sysfs_hprint(bucket_size,               bucket_bytes(c));
 531         sysfs_hprint(block_size,                block_bytes(c));
 532         sysfs_print(tree_depth,                 c->root->level);
 533         sysfs_print(root_usage_percent,         bch_root_usage(c));
 534
 535         sysfs_hprint(btree_cache_size,          bch_cache_size(c));
 536         sysfs_print(btree_cache_max_chain,      bch_cache_max_chain(c));
 537         sysfs_print(cache_available_percent,    100 - c->gc_stats.in_use);
 538
 539         sysfs_print_time_stats(&c->btree_gc_time,       btree_gc, sec, ms);
 540         sysfs_print_time_stats(&c->btree_split_time,    btree_split, sec, us);
 541         sysfs_print_time_stats(&c->sort.time,           btree_sort, ms, us);
 542         sysfs_print_time_stats(&c->btree_read_time,     btree_read, ms, us);
 543
 544         sysfs_print(btree_used_percent, bch_btree_used(c));
 545         sysfs_print(btree_nodes,        c->gc_stats.nodes);
 546         sysfs_hprint(average_key_size,  bch_average_key_size(c));
 547
 548         sysfs_print(cache_read_races,
 549                     atomic_long_read(&c->cache_read_races));
 550
 551         sysfs_print(writeback_keys_done,
 552                     atomic_long_read(&c->writeback_keys_done));
 553         sysfs_print(writeback_keys_failed,
 554                     atomic_long_read(&c->writeback_keys_failed));
 555
 556         if (attr == &sysfs_errors)
 557                 return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
 558                                                c->on_error);
 559
 560         /* See count_io_errors for why 88 */
 561         sysfs_print(io_error_halflife,  c->error_decay * 88);
 562         sysfs_print(io_error_limit,     c->error_limit >> IO_ERROR_SHIFT);
 563
 564         sysfs_hprint(congested,
 565                      ((uint64_t) bch_get_congested(c)) << 9);
 566         sysfs_print(congested_read_threshold_us,
 567                     c->congested_read_threshold_us);
 568         sysfs_print(congested_write_threshold_us,
 569                     c->congested_write_threshold_us);
 570
 571         sysfs_print(active_journal_entries,     fifo_used(&c->journal.pin));
 572         sysfs_printf(verify,                    "%i", c->verify);
 573         sysfs_printf(key_merging_disabled,      "%i", c->key_merging_disabled);
 574         sysfs_printf(expensive_debug_checks,
 575                      "%i", c->expensive_debug_checks);
 576         sysfs_printf(gc_always_rewrite,         "%i", c->gc_always_rewrite);
 577         sysfs_printf(btree_shrinker_disabled,   "%i", c->shrinker_disabled);
 578         sysfs_printf(copy_gc_enabled,           "%i", c->copy_gc_enabled);
 579
 580         if (attr == &sysfs_bset_tree_stats)
 581                 return bch_bset_print_stats(c, buf);
 582
 583         return 0;
 584 }
 585 SHOW_LOCKED(bch_cache_set)
 586
 587 STORE(__bch_cache_set)
 588 {
 589         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 590
 591         if (attr == &sysfs_unregister)
 592                 bch_cache_set_unregister(c);
 593
 594         if (attr == &sysfs_stop)
 595                 bch_cache_set_stop(c);
 596
 597         if (attr == &sysfs_synchronous) {
 598                 bool sync = strtoul_or_return(buf);
 599
 600                 if (sync != CACHE_SYNC(&c->sb)) {
 601                         SET_CACHE_SYNC(&c->sb, sync);
 602                         bcache_write_super(c);
 603                 }
 604         }
 605
 606         if (attr == &sysfs_flash_vol_create) {
 607                 int r;
 608                 uint64_t v;
 609                 strtoi_h_or_return(buf, v);
 610
 611                 r = bch_flash_dev_create(c, v);
 612                 if (r)
 613                         return r;
 614         }
 615
 616         if (attr == &sysfs_clear_stats) {
 617                 atomic_long_set(&c->writeback_keys_done,        0);
 618                 atomic_long_set(&c->writeback_keys_failed,      0);
 619
 620                 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
 621                 bch_cache_accounting_clear(&c->accounting);
 622         }
 623
 624         if (attr == &sysfs_trigger_gc) {
 625                 /*
 626                  * Garbage collection thread only works when sectors_to_gc < 0,
 627                  * when users write to sysfs entry trigger_gc, most of time
 628                  * they want to forcibly triger gargage collection. Here -1 is
 629                  * set to c->sectors_to_gc, to make gc_should_run() give a
 630                  * chance to permit gc thread to run. "give a chance" means
 631                  * before going into gc_should_run(), there is still chance
 632                  * that c->sectors_to_gc being set to other positive value. So
 633                  * writing sysfs entry trigger_gc won't always make sure gc
 634                  * thread takes effect.
 635                  */
 636                 atomic_set(&c->sectors_to_gc, -1);
 637                 wake_up_gc(c);
 638         }
 639
 640         if (attr == &sysfs_prune_cache) {
 641                 struct shrink_control sc;
 642                 sc.gfp_mask = GFP_KERNEL;
 643                 sc.nr_to_scan = strtoul_or_return(buf);
 644                 c->shrink.scan_objects(&c->shrink, &sc);
 645         }
 646
 647         sysfs_strtoul(congested_read_threshold_us,
 648                       c->congested_read_threshold_us);
 649         sysfs_strtoul(congested_write_threshold_us,
 650                       c->congested_write_threshold_us);
 651
 652         if (attr == &sysfs_errors) {
 653                 ssize_t v = bch_read_string_list(buf, error_actions);
 654
 655                 if (v < 0)
 656                         return v;
 657
 658                 c->on_error = v;
 659         }
 660
 661         if (attr == &sysfs_io_error_limit)
 662                 c->error_limit = strtoul_or_return(buf) << IO_ERROR_SHIFT;
 663
 664         /* See count_io_errors() for why 88 */
 665         if (attr == &sysfs_io_error_halflife) {
 666                 unsigned long v = 0;
 667                 ssize_t ret;
 668
 669                 ret = strtoul_safe_clamp(buf, v, 0, UINT_MAX);
 670                 if (!ret) {
 671                         c->error_decay = v / 88;
 672                         return size;
 673                 }
 674                 return ret;
 675         }
 676
 677         sysfs_strtoul(journal_delay_ms,         c->journal_delay_ms);
 678         sysfs_strtoul(verify,                   c->verify);
 679         sysfs_strtoul(key_merging_disabled,     c->key_merging_disabled);
 680         sysfs_strtoul(expensive_debug_checks,   c->expensive_debug_checks);
 681         sysfs_strtoul(gc_always_rewrite,        c->gc_always_rewrite);
 682         sysfs_strtoul(btree_shrinker_disabled,  c->shrinker_disabled);
 683         sysfs_strtoul(copy_gc_enabled,          c->copy_gc_enabled);
 684
 685         return size;
 686 }
 687 STORE_LOCKED(bch_cache_set)
 688
 689 SHOW(bch_cache_set_internal)
 690 {
 691         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 692         return bch_cache_set_show(&c->kobj, attr, buf);
 693 }
 694
 695 STORE(bch_cache_set_internal)
 696 {
 697         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 698         return bch_cache_set_store(&c->kobj, attr, buf, size);
 699 }
 700
 701 static void bch_cache_set_internal_release(struct kobject *k)
 702 {
 703 }
 704
 705 static struct attribute *bch_cache_set_files[] = {
 706         &sysfs_unregister,
 707         &sysfs_stop,
 708         &sysfs_synchronous,
 709         &sysfs_journal_delay_ms,
 710         &sysfs_flash_vol_create,
 711
 712         &sysfs_bucket_size,
 713         &sysfs_block_size,
 714         &sysfs_tree_depth,
 715         &sysfs_root_usage_percent,
 716         &sysfs_btree_cache_size,
 717         &sysfs_cache_available_percent,
 718
 719         &sysfs_average_key_size,
 720
 721         &sysfs_errors,
 722         &sysfs_io_error_limit,
 723         &sysfs_io_error_halflife,
 724         &sysfs_congested,
 725         &sysfs_congested_read_threshold_us,
 726         &sysfs_congested_write_threshold_us,
 727         &sysfs_clear_stats,
 728         NULL
 729 };
 730 KTYPE(bch_cache_set);
 731
 732 static struct attribute *bch_cache_set_internal_files[] = {
 733         &sysfs_active_journal_entries,
 734
 735         sysfs_time_stats_attribute_list(btree_gc, sec, ms)
 736         sysfs_time_stats_attribute_list(btree_split, sec, us)
 737         sysfs_time_stats_attribute_list(btree_sort, ms, us)
 738         sysfs_time_stats_attribute_list(btree_read, ms, us)
 739
 740         &sysfs_btree_nodes,
 741         &sysfs_btree_used_percent,
 742         &sysfs_btree_cache_max_chain,
 743
 744         &sysfs_bset_tree_stats,
 745         &sysfs_cache_read_races,
 746         &sysfs_writeback_keys_done,
 747         &sysfs_writeback_keys_failed,
 748
 749         &sysfs_trigger_gc,
 750         &sysfs_prune_cache,
 751 #ifdef CONFIG_BCACHE_DEBUG
 752         &sysfs_verify,
 753         &sysfs_key_merging_disabled,
 754         &sysfs_expensive_debug_checks,
 755 #endif
 756         &sysfs_gc_always_rewrite,
 757         &sysfs_btree_shrinker_disabled,
 758         &sysfs_copy_gc_enabled,
 759         NULL
 760 };
 761 KTYPE(bch_cache_set_internal);
 762
 763 SHOW(__bch_cache)
 764 {
 765         struct cache *ca = container_of(kobj, struct cache, kobj);
 766
 767         sysfs_hprint(bucket_size,       bucket_bytes(ca));
 768         sysfs_hprint(block_size,        block_bytes(ca));
 769         sysfs_print(nbuckets,           ca->sb.nbuckets);
 770         sysfs_print(discard,            ca->discard);
 771         sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
 772         sysfs_hprint(btree_written,
 773                      atomic_long_read(&ca->btree_sectors_written) << 9);
 774         sysfs_hprint(metadata_written,
 775                      (atomic_long_read(&ca->meta_sectors_written) +
 776                       atomic_long_read(&ca->btree_sectors_written)) << 9);
 777
 778         sysfs_print(io_errors,
 779                     atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
 780
 781         if (attr == &sysfs_cache_replacement_policy)
 782                 return bch_snprint_string_list(buf, PAGE_SIZE,
 783                                                cache_replacement_policies,
 784                                                CACHE_REPLACEMENT(&ca->sb));
 785
 786         if (attr == &sysfs_priority_stats) {
 787                 int cmp(const void *l, const void *r)
 788                 {       return *((uint16_t *) r) - *((uint16_t *) l); }
 789
 790                 struct bucket *b;
 791                 size_t n = ca->sb.nbuckets, i;
 792                 size_t unused = 0, available = 0, dirty = 0, meta = 0;
 793                 uint64_t sum = 0;
 794                 /* Compute 31 quantiles */
 795                 uint16_t q[31], *p, *cached;
 796                 ssize_t ret;
 797
 798                 cached = p = vmalloc(ca->sb.nbuckets * sizeof(uint16_t));
 799                 if (!p)
 800                         return -ENOMEM;
 801
 802                 mutex_lock(&ca->set->bucket_lock);
 803                 for_each_bucket(b, ca) {
 804                         if (!GC_SECTORS_USED(b))
 805                                 unused++;
 806                         if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
 807                                 available++;
 808                         if (GC_MARK(b) == GC_MARK_DIRTY)
 809                                 dirty++;
 810                         if (GC_MARK(b) == GC_MARK_METADATA)
 811                                 meta++;
 812                 }
 813
 814                 for (i = ca->sb.first_bucket; i < n; i++)
 815                         p[i] = ca->buckets[i].prio;
 816                 mutex_unlock(&ca->set->bucket_lock);
 817
 818                 sort(p, n, sizeof(uint16_t), cmp, NULL);
 819
 820                 while (n &&
 821                        !cached[n - 1])
 822                         --n;
 823
 824                 unused = ca->sb.nbuckets - n;
 825
 826                 while (cached < p + n &&
 827                        *cached == BTREE_PRIO)
 828                         cached++, n--;
 829
 830                 for (i = 0; i < n; i++)
 831                         sum += INITIAL_PRIO - cached[i];
 832
 833                 if (n)
 834                         do_div(sum, n);
 835
 836                 for (i = 0; i < ARRAY_SIZE(q); i++)
 837                         q[i] = INITIAL_PRIO - cached[n * (i + 1) /
 838                                 (ARRAY_SIZE(q) + 1)];
 839
 840                 vfree(p);
 841
 842                 ret = scnprintf(buf, PAGE_SIZE,
 843                                 "Unused:                %zu%%\n"
 844                                 "Clean:         %zu%%\n"
 845                                 "Dirty:         %zu%%\n"
 846                                 "Metadata:      %zu%%\n"
 847                                 "Average:       %llu\n"
 848                                 "Sectors per Q: %zu\n"
 849                                 "Quantiles:     [",
 850                                 unused * 100 / (size_t) ca->sb.nbuckets,
 851                                 available * 100 / (size_t) ca->sb.nbuckets,
 852                                 dirty * 100 / (size_t) ca->sb.nbuckets,
 853                                 meta * 100 / (size_t) ca->sb.nbuckets, sum,
 854                                 n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
 855
 856                 for (i = 0; i < ARRAY_SIZE(q); i++)
 857                         ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 858                                          "%u ", q[i]);
 859                 ret--;
 860
 861                 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
 862
 863                 return ret;
 864         }
 865
 866         return 0;
 867 }
 868 SHOW_LOCKED(bch_cache)
 869
 870 STORE(__bch_cache)
 871 {
 872         struct cache *ca = container_of(kobj, struct cache, kobj);
 873
 874         if (attr == &sysfs_discard) {
 875                 bool v = strtoul_or_return(buf);
 876
 877                 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
 878                         ca->discard = v;
 879
 880                 if (v != CACHE_DISCARD(&ca->sb)) {
 881                         SET_CACHE_DISCARD(&ca->sb, v);
 882                         bcache_write_super(ca->set);
 883                 }
 884         }
 885
 886         if (attr == &sysfs_cache_replacement_policy) {
 887                 ssize_t v = bch_read_string_list(buf, cache_replacement_policies);
 888
 889                 if (v < 0)
 890                         return v;
 891
 892                 if ((unsigned) v != CACHE_REPLACEMENT(&ca->sb)) {
 893                         mutex_lock(&ca->set->bucket_lock);
 894                         SET_CACHE_REPLACEMENT(&ca->sb, v);
 895                         mutex_unlock(&ca->set->bucket_lock);
 896
 897                         bcache_write_super(ca->set);
 898                 }
 899         }
 900
 901         if (attr == &sysfs_clear_stats) {
 902                 atomic_long_set(&ca->sectors_written, 0);
 903                 atomic_long_set(&ca->btree_sectors_written, 0);
 904                 atomic_long_set(&ca->meta_sectors_written, 0);
 905                 atomic_set(&ca->io_count, 0);
 906                 atomic_set(&ca->io_errors, 0);
 907         }
 908
 909         return size;
 910 }
 911 STORE_LOCKED(bch_cache)
 912
 913 static struct attribute *bch_cache_files[] = {
 914         &sysfs_bucket_size,
 915         &sysfs_block_size,
 916         &sysfs_nbuckets,
 917         &sysfs_priority_stats,
 918         &sysfs_discard,
 919         &sysfs_written,
 920         &sysfs_btree_written,
 921         &sysfs_metadata_written,
 922         &sysfs_io_errors,
 923         &sysfs_clear_stats,
 924         &sysfs_cache_replacement_policy,
 925         NULL
 926 };
 927 KTYPE(bch_cache);